Parallel image thinning through topological operators on shared memory parallel machines

📝 Abstract

In this paper, we present a concurrent implementation of a powerful topological thinning operator. This operator is able to act directly over grayscale images without modifying their topology. We introduce an adapted parallelization methodology which combines split, distribute and merge (SDM) strategy and mixed parallelism techniques (data and thread parallelism). The introduced strategy allows efficient parallelization of a large class of topological operators including, mainly, {\lambda}-leveling, skeletonization and crest restoring algorithms. To achieve a good speedup, we cared about coordination of threads. Distributed work during thinning process is done by a variable number of threads. Tests on 2D grayscale image (512*512), using shared memory parallel machine (SMPM) with 8 CPU cores (2x Xeon E5405 running at frequency of 2 GHz), showed an enhancement of 6.2 with a maximum achieved cadency of 125 images/s using 8 threads.

💡 Analysis

In this paper, we present a concurrent implementation of a powerful topological thinning operator. This operator is able to act directly over grayscale images without modifying their topology. We introduce an adapted parallelization methodology which combines split, distribute and merge (SDM) strategy and mixed parallelism techniques (data and thread parallelism). The introduced strategy allows efficient parallelization of a large class of topological operators including, mainly, {\lambda}-leveling, skeletonization and crest restoring algorithms. To achieve a good speedup, we cared about coordination of threads. Distributed work during thinning process is done by a variable number of threads. Tests on 2D grayscale image (512*512), using shared memory parallel machine (SMPM) with 8 CPU cores (2x Xeon E5405 running at frequency of 2 GHz), showed an enhancement of 6.2 with a maximum achieved cadency of 125 images/s using 8 threads.

📄 Content

Parallel Image Thinning Through Topological Operators
On Shared Memory Parallel Machines

Ramzi MAHMOUDI1, Mohamed AKIL1, Petr MATAS1,2

1Université Paris-Est, Laboratoire d’Informatique Gaspard-Monge, Equipe A3SI,
ESIEE Paris Cité Descartes, BP99, 93162 Noisy Le Grand, France

2Department of Applied Electronics and Telecommunications, University of West Bohemia, Univerzitní 26, 306 14 Plzeň, Czech Republic

{mahmoudr, akilm, matasp}@esiee.fr

Abstract

In this paper, we present a concurrent implementation of a powerful topological thinning operator. This operator is able to act directly over grayscale images without modifying their topology. We introduce an adapted parallelization methodology which combines split, distribute and merge (SDM) strategy and mixed parallelism techniques (data and thread parallelism). The introduced strategy allows efficient parallelization of a large class of topological operators including, mainly,
-leveling, skeletonization and crest restoring algorithms. To achieve a good speedup, we cared about coordination of threads. Distributed work during thinning process is done by a variable number of threads. Tests on 2D grayscale image (512*512), using shared memory parallel machine (SMPM) with 8 CPU cores (2× Xeon E5405 running at frequency of 2 GHz), showed an enhancement of 6.2 with a maximum achieved cadency of 125 images/s using 8 threads.

1. Introduction

   In many computer vision applications, standard techniques of pattern recognition are thinning algorithms. As a preprocessing stage, these algorithms have been used for the recognition of handwriting or printed characters, fingerprints, chromosomes and biological cell structures, etc. [1]. Topological thinning and skeletonization are ones of the most cardinal operators for this kind of preprocessing, especially since the development, by our team, of an efficient thinning algorithm able to act directly over grayscale image [2]. Using topological operators allows topology preservation which results in conservation of important significant information [3]. This conservation was impossible in the case of binary image processing [4]. Early thinning algorithms were designed for serial implementations, but since parallel computers are available several approaches have been developed with parallel processing [5,6]. In [5], Heydorn presents a concept for an implementation of different parallel thinning algorithms on parallel processors. The emphasis is put on a good parallelization using fine granularity and the simultaneous usage of vectorization.

This paper describes an adapted parallelization methodology combining split, distribute and merge (SDM) strategy based upon the well-known principle of divide-and-conquer and thread coordination which allows an efficient parallelism for introduced thinning operator on shared memory machines. Proposed strategy can also be applied for any topological operator having the same characteristics based on elementary operations of point characterization and similar algorithmic structure as we will demonstrate later.

This paper is organized as follows: in section 2, some basic notions of topological operators are summarized; the original algorithm of thinning is introduced. We define also the class of operators that our parallelization strategy may cover. In section 3, parallelization strategy, that has been adopted, is introduced. In section 4, Thread coordination and synchronization is discussed. In section 5, as a concrete example of introduced strategy and threads synchronization techniques, a parallel version of thinning algorithm is presented. Experimental analyzes results of different implementations are also presented and discussed. Finally, we conclude with summary and future work in section 6.
2. Topological thinning operator

Skeletonization and thinning are major applications of topology in image processing. A great number of thinning algorithms for binary images have been developped [7]. The use of this kind of images assumes a prior segmentation which implies a loss of information. Some attention has been given to the development of thinning algorithms acting directly over grayscale images. Dyer and Rosenfeld [8] proposed an algorithm based on a weighted notion of connectedness. The thinning is done directly over the graylevels of the points but, as the authors showed, the connectivity of objects is not always preserved. Other works [9] use an implicit image binarization into a background and a graylevel foreground. The graylevel information guides the removal of points of the foreground that are simple, in the binary sense. This technique makes it possible to obtain certain desired geometric properties. Inspired by this technique, M. Couprie and G. Bertrand [2] propose a filtered thinning method that allows

This content is AI-processed based on ArXiv data.